Search

Mixed-Linkage (1->3),(1->4)-ß-D-Glucan (MLG) is the predominant cell wall component of cereal grain endosperm. MLG is unbranched and unsubstituted chain of D-Glucose. It has two types of linkages between D-Glucoses: (1->3) and (1->4). Two proteins have been identified to be able to synthesize MLG: cellulose synthase-like F6 and H1 (CSLF6 and CSLH1). However, several questions remain to be answered. First, the subcellular localization of the biosynthesis of MLG has been a much-debated topic. Evidences exist that support both Golgi- and plasma membrane (PM)-localization. In addition, whether a protein complex is needed and what proteins interact to synthesize MLG are still unknown.The first chapter addresses the localization and protein-protein interactions of barley CSLF6 and CSLH1 using Agro-mediated transient expression in tobacco and bimolecular fluorescence complementation (BiFC). BiFC showed that MLG biosynthesis requires protein complexes that contain two or more CSLF6 or CSLH1 proteins. It was shown that CSLF6 and CSLH1 can form homomeric and heteromeric complexes. In addition, the subcellular localization was consistent with the previous report that MLG was synthesized in the PM. CSLF6 showed a multiple-cell organelle localization pattern including Golgi (rarely), endoplasmic reticulum (ER), and PM, while CSLH1 was detected mostly in the ER. The MLG synthase activity was confirmed for CSLF6, while CSLH1 showed no detectable activity.To further reveal other unknown proteins that may be involved in MLG biosynthesis, a bioinformatics-based study is presented in chapter 2. A barley global gene-co-expression network (GCN) was constructed and optimized. A list of potential co-expressors and transcription factors for CslF6 was identified using the GCN. Five hundred of RNA-Seq libraries were acquired and processed. The optimized GCN was built after a systematic evaluation of the normalization and network inference methods. The GCN was used to infer for barley CslF6 the co-expressed genes and transcription factors.The third study aimed to alleviate some of the drawbacks of traditional method used in building GCNs (which was used in the second study) by constructing GCNs using deep neural networks (DNNs). The results show that DNN can be used to construct GCNs that not only have strong ability to integrate prior knowledge of know co-expressions, but also generalize well to infer for unknown ones. Architectures and hyperparameters were explored. The deep neural network method was verified by comparing the area under the receiver operating characteristic curve and area under the precision recall curve of two test sets. The final GCN showed similar performances to the traditional methods for inferring gene functions, but had significant improvement for inferring genes that are in the same pathway, which helped reveal the proteins that may be involved in MLG biosynthesis. A list of co-expressed genes and transcription factors was obtained and compared to the results from chapter 2.